Signature and specification matching

Large libraries of software components hold great potential as a resource for software engineers, but to utilize them fully, we need to be able: (1) to locate components in the library; (2) to organize the library in a way that facilitates browsing and improves e ciency of retrieval; and (3) to compare the description of a library component to the description of what we want. A key requirement in all of these problems is to be able to compare two software components to see whether they match. In this dissertation, we consider two di erent kinds of semantic descriptions of components to determine whether components match: signatures (type information) and speci cations (behavioral information). Semantic descriptions o er advantages over either textual descriptions, such as variable names, or structural descriptions, such as control ow graphs. Using semantic information focuses on what the components do rather than how they do it. Signatures and speci cations are natural ways of describing software components and have well-understood properties, such as type equivalence and logical relations between formal speci cations, that enable us both to de ne matches precisely and to automate the match. This dissertation makes the following contributions: Foundational. Within a general, highly modular, and extensible framework, we de ne matching for two kinds of semantic information (signatures and speci cations) and two granularities of components (functions and modules). Each kind of matching has a generic form, within which all of the matches are related and may, in some cases, be composed. The orthogonality of the matches allows us to de ne match on modules independently of the particular match used on functions in the modules. Applications. We show how the de nitions of matching can be applied to the problems of retrieval from libraries, indexing libraries, and reuse of components. We demonstrate the various signature and speci cation matches with examples of typical uses in each application. Engineering. We describe our implementations of function and module signature match, function speci cation match, function signature-based indexing, and function signaturebased retrieval. These implementations demonstrate the feasibility of our approach and allow us to illustrate the applications with results from a moderately-sized component library.